THEMA Working Paper n°2020-03 CY Cergy Paris Université, France
Diagonal Cumulation and Sourcing Decisions
Pamela Bombarda, Elisa Gamberoni
April 2020
Diagonal Cumulation and Sourcing Decisions ∗
Pamela Bombarda
†and Elisa Gamberoni
‡April 25, 2020
Abstract
Free trade agreements (FTAs) are characterized by rules of origin (RoO) and cumula- tion. These rules define which intermediate goods allow a final product to qualify for preferential access. Recent literature shows that RoO led to a reduction in imports of intermediate goods from third countries relative to partners. We consider the impact of the Pan-European Cumulation System (PECS), which provided the possibility of cumulating stages of production across European Union’s FTA peripheral partners.
We find that PECS reshaped regional supply chains by increasing imports of inter- mediates among these peripheral countries relative to both the European Union (EU) and third countries. We also find that PECS reinforced their value chain links with third countries relative to the EU, contributing to multilateralize regionalism.
JEL classification: F12, F13, F14, F15.
Keywords: Intermediate Trade, rules of origin, diagonal cumulation, PECS, input- output tables.
∗Acknowledgments: We would like to thank Carlo Altomonte, Andy Bernard and Paola Conconi for their insightful comments and discussions on this topic. Remarks from participants at the SETC 2019, ETSG Bern, and Cergy seminar contributed to improving this paper. For their suggestions and comments, we would also like to thank Arup Banerji, Donato De Rosa, Vivek Suri, and Gallina A. Vincelette. All errors are ours. Bombarda thanks support from the Labex MME-DII program (ANR-11-LBX-0023-01). The views, findings, interpretations, and conclusions expressed in this paper are entirely those of the authors.
They do not necessarily represent the view of the World Bank, its Executive Directors or the countries they represent.
†CY Cergy Paris Universit´e, CNRS, THEMA, F-95000 Cergy, France, [email protected].
‡World Bank.
1 Introduction
Global Value Chains (GVCs), as a way to organize the production process across nations, have large implications on international trade flows. Throughout 1995 and 2011 trade within GVCs accounted for 60–67 percent of global trade in value-added terms. In 2009, 66 percent of world exports was characterized as trade flows in intermediate goods. This is suggestive of a production process split across nations. Statistics on intra-regional exports of trade in intermediate and final manufactured goods over 1995–2015 for Europe, the Americas, Asia, and the rest of the world highlight the large shares of intra-regional linkages (World Bank and the WTO, 2017). This confirms that GVCs are organized mainly at the regional level, creating what Baldwin and Lopez-Gonzalez (2013) call “Factory Asia,” “Factory North America,” and “Factory Europe.”
The trade literature has analyzed the determinants behind the participation of countries in GVCs. This literature has focused on countries’ endowments and institutions. Antr`as and Helpman(2008) propose a model in which contractual frictions affect the decision to offshore versus outsource part of the production process. Nunn (2007) provides empirical evidence about the role of the quality of contract-related institutions on comparative advantage in contract-intensive goods. Moreover, Antr`as et al. (2012) show that the distance of the export basket to final use (a measure of upstreamness in the production process) negatively correlates with a country’s quality of institutions, skill endowment, and access to finance.
The literature has also suggested that location and trade related costs matter for GVCs.
For instance, Antr`as and de Gortari (2017) show that the optimal location of production of a given stage in the global value chain depends also on the proximity of that location to the preceding and the subsequent desired locations of production. They show that it is optimal to locate relatively downstream stages of production in relatively central locations.
To explain the ability of a country to participate in GVCs, this paper considers the role of a specific European Union (EU) trade policy: the introduction in 1997 of the Pan European Cumulation System (henceforth PECS). PECS aimed at harmonizing the heterogeneous set of free trade agreements (FTAs) that the EU had put in place with Central and Eastern European countries (CEECs) in the early nineteens.1 At the heart of each FTAs there are rules of origin (henceforth RoO), and their corresponding cumulation of origin rules. By determining the origin of a product, RoO define whether or not a good qualifies for preferen- tial access. Cumulation of origin rules defines whether a firm can use imported intermediate goods from a specific country, so that the final product of the importing firm does not lose the originating status. A preferential regime can allow either for bilateral, diagonal, or full cumulation. Bilateral cumulation is operated between two partners. This implies that pro- ducers in either partner country can use intermediates originating in the other’s country as if they are originated in their own country. Operations carried out in one partner country can be aggregated with the operations carried out in another partner country to confer orig-
1CEECs is an OECD term that includes BAFTA and CEFTA countries. BAFTA includes Estonia,
Latvia, Lithuania. CEFTA includes: Bulgaria, the Czech Republic, Hungary, Poland, Romania, Slovak Republic and Slovenia.
inating status to goods traded between them. Diagonal cumulation operates between more than two countries. Considering three countries, A, B and C, diagonal cumulation requires these countries to be bound together by bilateral FTAs characterized by identical set of RoO. Then, a firm in country A can use intermediate goods originating from FTA partners in B and C to produce an originating product in country A. Alternatively, a preferential regime can allow for full cumulation. In this case all stages of production coming from free trade area partners can be counted as qualifying to achieve the origin status, regardless of whether the processing is sufficient to confer originating status.
The proliferation of different FTAs in the early ’90s generated an increasing level of complexity among countries participating or willing to participate into the slicing up of the value added chain. The set of FTAs signed in the European region were all characterized by bilateral cumulation. Let’s consider for example the bilateral FTAs that the EU signed with Hungary and Poland in the 1990s. Suppose that RoO on cloth impose that all shirts imported duty-free into the EU market to be made either of EU cloth or of locally-produced cloth. RoO will force Hungarian shirt producers to switch from buying Polish cloth to buying EU cloth in order to get preferential treatment for their exports of shirts in the EU market. The bilateral cumulation plus RoO act like a Hungarian tariff on Polish cloth.2 Thus, bilateral cumulation was limiting the fragmentation of the production process within the area, specifically among the EU and CEECs.
To tame the European “Spaghetti Bowl”, the EU pushed through PECS. PECS, which came in 1997, completed the missing FTAs, it harmonized the rules of origin protocols of all the underlying FTAs, and most importantly introduced diagonal cumulation. With diagonal cumulation, countries belonging to the PECS agreement can source parts and components from within PECS partners without fear of the resulting product losing its origin status, and thus its right to preferential treatment. Using the above example, when cumulation is diagonal this would allow the Hungarian shirt-makers to use Polish cloth in meeting the RoO (Baldwin,2013). PECS provided an opportunity to split the value chain of production among several PECS peripheral partners, since the final good could still benefit from the preferential tariffs.3 As a result, countries belonging to the cumulation zone enjoyed a cost advantage, which influenced the organization of value chains for goods destined to the EU market. Since the cumulation zone was initially established among countries in the region, it should have contributed to the creation of a “factory Europe.”
Using PECS as a natural experiment, this paper studies how the relaxation of RoO achieved via diagonal cumulation altered the setting up of international supply chains among peripheral countries in the PECS network. We consider changes in imports of intermedi- ate goods over the period 1995-2002 in different PECS peripheral countries, i.e. BAFTA and CEFTA members. Specifically, we compare imports coming from different exporting countries: the other Spokes (PECS network), third countries (also called rest of the world countries, RoW), and EU15.4 Our findings show that, as a result of diagonal cumulation,
2This example is discussed inBaldwin(2013).
3PECS peripheral partners are also referred to as Spoke countries.
4EU15 is used to indicate those countries that joined the EU until 1995.
PECS peripheral countries increase imports of intermediate goods among themselves rela- tive to other countries (RoW and EU15), and that this increase is larger, the stricter the rules of origin applied to the related final products. Moreover, we show that following di- agonal cumulation, PECS peripheral partners also increase their imports of intermediate goods from third countries (RoW) compared to the EU15. The evidence presented in this paper suggests that diagonal cumulation have led to a reassessment of sourcing decisions established during the pre-PECS period. Indeed, diagonal cumulation allows preferential access for exports of final goods that are produced with intermediates imported by a larger set of countries. Since RoO do not typically require whole obtained originating status, the increase in sourcing choices may have led PECS peripheral countries to import more inter- mediate goods also from RoW. Therefore, our results also support the idea that diagonal cumulation can lead to a multilateralization of regionalism (Baldwin, 2006).
Our analysis is developed in three steps. First, we present some stylized facts on the extent to which EU-RoO on final goods affect intermediate goods. We show that the re- strictiveness of RoO is linked to the nature of the production process in which intermediate goods participate. For example, intermediate goods classified as textiles mainly enter into the production process of textile goods. Final goods classified as textiles require more than 40 percent of regional value added to benefit from preferential access in the EU. Intermedi- ate goods classified as vegetables, by contrast, enter mainly into the production process of final goods classified as chemicals, which need to fulfill less restrictive rules of origin. This evidence stresses the need for translating in intermediate goods terms RoO defined at the final good level.
Second, we followConconi et al. (2018) and use a triple difference model to analyze the impact of diagonal cumulation on changes in each peripheral country’s imports of intermedi- ates before and after PECS (1995 - 2002). This allows us to control for unobservable product specific time-invariant characteristics, as well as product level trends common across our set of countries. As the effect of diagonal cumulation is strictly linked to the restrictiveness of rules of origin, we exploit an index developed by Cadot et al. (2006). This index mea- sures the restrictiveness of RoO applied to final goods. To evaluate the effects of PECS on imports of intermediate goods, we translate this RoO index in terms of intermediate goods by combining it with the US IO1997 table. In the empirical strategy, we compare changes in Spokes’ imports of intermediates from the RoW to changes in Spokes’ imports from the other Spokes. We show that following the introduction of diagonal cumulation the intermediate imports from the RoW declined compared to imports from the other Spokes.
The magnitude of this decline is positively associated with the extent of the restrictive RoO applied to the associated final goods.
Finally, owing to the nature of the experiment, we also exploit other control groups and try to disentangle trade creation from trade diversion effects in intermediate goods.
Switching from bilateral to diagonal cumulation can boost trade in intermediates thanks to the fact that diagonal cumulation provides new sourcing possibilities compared to the initial FTAs characterized by more stringent (bilateral) rules of cumulation. Thus, diagonal cumulation could lead Spoke countries to change their supply chain linkages, by substituting
intermediates imported from the EU market with cheaper intermediates coming from Spoke countries. As a consequence of this, also exports from third countries, outside the PECS cumulation zone, may have been favored. The relaxation of RoO induced by PECS, may have allowed Spoke countries to increase their imports of intermediates from third countries (RoW), while still enjoying preferential treatment when exporting to EU market. In sum, diagonal cumulation may have led exporting firms in Spoke countries to reassess sourcing decisions towards other Spoke as well as RoW countries, and away from the EU.
To assess these additional mechanisms, we consider two alternative changes in Spokes’
imports. First, we compare changes in imports of intermediates by each Spoke country from the RoW to those from the EU15. Our results show that diagonal cumulation reduced imports of intermediate goods from the EU15 vis-`a-vis RoW. Second, we consider changes in imports by each Spoke country from the PECS network to changes in imports from the EU15. Similarly, we find that diagonal cumulation reduced imports of intermediate goods from the EU15 vis-`a-vis the other Spokes. These additional results highlight that Spokes countries reorganize their production process towards the other peripheral countries in the PECS network as well as third countries, moving away from EU15. The magnitude of these effects is also positively associated with the extent of the restrictive RoO applied to the associated final goods.
Our findings suggest that indeed PECS allowed a reassessment of sourcing decisions.
Diagonal cumulation allowed peripheral countries to re-organize global value chain links. In terms of magnitude, our estimates imply that PECS reduced Spokes’ intermediate imports from RoW relative to the PECS network by around 14%. Using the other control groups, we also show that PECS increased imports from RoW and PECS network relative to the EU15 respectively by 3% and 9%. Due to the specific economic integration in the European region, we try to address possible identification threats that may arise from pre-existing trends. Finally, we provide several robustness checks to show that our results are robust to using different samples of exporting countries, using an alternative way to construct the dependent variable, and the control variable (where we focus on RoO based on whether final goods face stringent value added requirements).
Related Literature
The theoretical literature on regional trade agreements and fragmentation of the production process is vast (Ornelas et al., 2019, Blanchard, 2015, Antr`as and Staiger, 2012 among others). A strand of this literature has shown that FTAs can potentially lower trade costs if the benefits from preferential access outweigh the costs of fulfilling RoO, therefore affecting a firm’s sourcing decisions. Demidova et al. (2012) build a heterogeneous firm setting which shows that firms sort according to the export markets and the different types of trade policy.
They model the response of Bangladesh firms in two sectors, the woven and non-woven sectors, with respect to the decision to export to the EU market under the “Everything But Arms” (EBA) initiative, and to the US market under the quota regime. In both cases firms can take advantage of these trade policies only if they comply with RoO. Modeling RoO as
an additional marginal and fixed cost, they show that firms that take advantage of the less restrictive EU RoO are less productive than those firms that export to the US, where tariffs are higher. By adding an intermediate good sector in a hub-spoke setting, Bombarda and Gamberoni (2013) show that only the most productive final good firms are able to export under preferential tariffs associated with RoO and bilateral cumulation. In their model RoO and diagonal cumulation affect sourcing decisions: switching from bilateral to diagonal cumulation relaxes the restrictiveness of RoO, leading to an increase in trade among firms in the Spoke countries. Our findings build on this theoretical literature and confirm the prediction that diagonal cumulation relaxes the restrictiveness of RoO by allowing larger trade among Spoke countries.
This paper further adds to recent empirical research that shows that RoO affect sourc- ing decisions. To the best of our knowledge, we provide first-time evidence that cumulation of origin rules affects input choices, and that less stringent rules may help reducing trade diversion effects in intermediate goods. Existing empirical work shows that RoO increase production costs and negatively affect trade flows. Anson et al. (2005) show that in the case of the North American Free Trade Agreement (NAFTA), up to 40 percent of Mex- ico’s preferential access to the US market in 2000 (estimated at 5 percent) was absorbed by RoO-related administrative costs. Cadot et al. (2006) construct, for both NAFTA and EU- related preferences, an index of restrictiveness of RoO at the six-digit level of the harmonized system. This index shows that RoO tend to be more restrictive for activities with greater processing, that sectors with higher RoO have lower utilization rates, and that non-least developing countries face restrictive RoO in sectors in which they have a revealed compar- ative advantage. Unlike this literature that considers the role of RoO on trade flows and utilization rates, our paper focus on RoO and sourcing decisions for CEECs countries as an example of the participation of EU peripheral countries in GVCs under a hub-spoke setting.
Closely related to out paper, is the work by Augier et al. (2005). Using aggregated data, they show that lack of cumulation before PECS impeded trade by 10% to 50%, depending on the time period and group of countries concerned. Differently from Augier et al.(2005), we account for the specific nature of RoO and their corresponding cumulation rules. Specif- ically, we work with data at the six digit level of the Harmonized System (HS6), which has about 5,000 product categories. Moreover, using a disaggregated IO table, we assess the effect of RoO on trade within global value chains rather than on overall trade.
Our paper also relates to the recent work measuring global value chains (Johnson and Noguera, 2017, Koopman et al., 2014 and Antr`as et al., 2012 among others). Specifically, we relate to studies evaluating the role of government policies on the ability of a country to participate in GVCs. More closely related to our work is the recent analysis byCaliendo and Parro (2015) and Conconi et al. (2018). Caliendo and Parro (2015) study the impact of NAFTA’s tariff reductions extending the Eaton and Kortum (2002) model to account for multiple-sector linkages. They find that the trade created, mostly between NAFTA members, was larger than the trade diverted from other economies. Conconi et al. (2018) consider the role of NAFTA RoO in affecting trade creation and diversion. They show that, controlling for the size of the preference margin, NAFTA RoO led to a sizeable reduction in
Mexico’s imports of intermediate goods from third countries relative to NAFTA partners.
For their analysis they built the RoO index using Input-Output tables. Specifically, for each intermediate good, they construct a measure based on the number of final goods facing a RoO-related restriction. They then use triple difference methodology to estimate the effect of NAFTA on non-members. Differently from Caliendo and Parro(2015) andConconi et al.(2018), we focus on diagonal cumulation, and show that more flexible RoO can revert sourcing decisions that may have resulted from the introduction of a FTA. Our results suggest that diagonal cumulation can foster trade among peripheral FTA members and may provide for an outcome closer to a multilateral liberalization. Focusing on EU-RoO allows us to take advantage of different control groups to better disentangle the effect of PECS on trade creation and diversion in intermediate goods.
The paper is structured as follows. Section2presents preliminary evidence on the role of PECS in influencing sourcing decisions. Section 3describes the data and variables construc- tion. Section 4 presents the empirical strategy, and the possible identification threats. The estimation results, and robustness checks are discussed in Section 5. Section 6concludes.
2 A Glance at PECS and Sourcing Decisions
FTAs have been a major instrument behind reciprocal liberalization. Between 1950 and 2017, 288 Regional Trade Agreements entered into force with FTAs representing the largest part of the category.5 To prevent trade deflection or simple transshipment— whereby prod- ucts from non-preferred countries are redirected through a free trade partner to avoid the payment of customs duties— RoO are used in FTAs. Theoretically, for each FTA, a par- ticular set of RoO can be established, negotiated according to the trade, industrial, and economic interests of each party.
Around 1993, trade in Europe was regulated by roughly 60 bilateral and plurilateral FTAs regulating European trade (Baldwin, 2013). These FTAs were characterized by bilateral cumulation, which implied that only inputs among the countries belonging to the FTA could be cumulated to count towards the establishment of the origin of the product. In the example discussed previously, the bilateral FTAs that the EU signed with Hungary and Poland in the 1990s foresaw bilateral cumulation. This implied that Hungary could cumulate inputs only from the EU or from itself to obtain origin status for the related final good, and export at preferential tariffs to the EU market (Baldwin, 2013). The existence of multiple FTAs led to a spaghetti bowl of preferential areas (and thus RoO) that did not favor the slicing up of the value chain across Europe. To reduce the heterogeneity of RoO in the pan-Euro-Mediterranean region, the EU and its partners agreed on a single set of rules of origin. This was achieved in 1997, which saw the creation of PECS. While the PECS system was introduced in 1997, countries did not join immediately. Among the BAFTA and
5For further information see http://rtais.wto.org/UI/PublicMaintainRTAHome.aspx.
CEFTA countries, the system could be considered mainly in place as of 1999.6
Since this paper analyzes the impact of RoO and diagonal cumulation on sourcing de- cision, we exploit the introduction of the PECS and investigate changes in imports among a group of the EU FTA’s peripheral partners, BAFTA and CEFTA (also called Spokes), between 1995 and 2002. To provide preliminary evidence of the effect of PECS on trade relationships, in Figure 1 we plot the evolution of imports of Spoke countries (in sample) before and after the introduction of PECS.7 The vertical axis measures changes in imports by each Spoke country from third countries (non-participating countries). The horizontal axis measures changes in overall imports by each Spoke country from the rest of the Spoke countries before and after 1997. Figure 1 shows that, between 1995 and 2002, the major- ity of Spoke countries experienced a larger increase in imports from the rest of the Spoke countries as opposed to the average imports from non-participating countries.
Figure 1: Change in Imports (1995–2002)
Source: Authors’ calculation using WITS Comtrade database
6The system was based on the EEA agreement (1994) between the EU, the EFTA (European Free
Trade Association), the CEEC (Central Eastern European Countries) and the Baltic States. Table 15 in AppendixAprovides further details about the preferential agreements providing for diagonal cumulation of origin between the EU and the Spoke countries. The system was then widened to Slovenia and to industrial products originating in Turkey (1999). The system was also enlarged to the Faroe Islands in 2005 and later to the countries of the Mediterranean and Balkan region.
7Our analysis considers imports of BAFTA and CEFTA countries. Notice that Bulgaria, the Slovak Republic, and Slovenia are dropped from our analysis due to absence of tariff data in the period under consideration. Table 9 in AppendixA provides a complete list of countries used in our study. Additional details about trade and tariff data used can be found in Section (3.2).
3 Data Description
This section describes the data used in the analysis and the methodology adopted to con- struct the variables of interest.
3.1 RoO Index at the Intermediate Level
We start the analysis by discussing the construction of the main variable of interest, which summarizes the difficulty in sourcing an intermediate product according to the RoO faced by the related final good. We follow Conconi et al.(2018) in constructing a data set that maps input-output linkages to the EU-RoO (which are defined according to the six digits of the 1992 Harmonized System). These input-output linkages are based on the US IO1997 table and are converted into the six digits of the 1988/1992 Harmonized System of classification, HS6 (see Conconi et al., 2018). Despite focusing on the EU context, we use the US IO tables for two reasons. Firstly, the match allows us to keep the same level of disaggregation of the trade data, that is HS6. Secondly, it enables us to follow the rationale in Rajan and Zingales (1998), who construct a measure of the extent of an industry’s dependence on external funds in the United States, and apply this measure to all other countries. As pointed out by the authors, the assumption is that “there is a technological reason why some industries depend more on external finance than others” and these technological reasons are the same in the United States as in other countries. Additionally, the authors use US data on capital markets since, among other things, these markets are among “the most advanced in the world, and large publicly traded firms typically face the least frictions in accessing finance. Thus, the amount of external finance used by large firms in the United States is likely to be a relatively pure measure of their demand for external finance.” In the same spirit, we exploit the US IO tables to better identify a production process that is less subject to distortions (such as product market restrictions), and therefore can be considered among the most efficient, allowing the identification of a production process driven by technological reasons.
Before describing the variable used to capture EU-RoO, it seems important to discuss the origin-determining criteria behind the rules of origin. These criteria determine how and when a product can be considered as originating in the European Community or in the partner country. Origin confers certain benefits on goods traded between countries that have agreed such an arrangement, usually entry of products at a reduced tariff rate. Therefore, origin criteria generally demand that these products undergo a certain amount of working or processing in the origin country. Specifically, a product is considered as originating if it has been wholly obtained, or sufficiently worked or processed with wholly or partly imported materials (Articles 5, 6, and 7 of the harmonized Protocol 4). Wholly obtained products are mainly mining, agricultural and fisheries products. While in the field of industrial products, most products are required to have undergone sufficient processing in either the Community or the partner country. There are three criteria used in determining sufficient working or processing: value added rules, change of tariff classification, and specific rules. Value
added rule means that the value of the non-originating materials must not exceed a certain percentage of the ex-works price of the finished product. Change of tariff classification (mostly at HS4) requires the non-originating raw materials or components used to have a different HS tariff classification from the HS tariff classification of the finished product.
Specific rules may require additional minimum processing permitted to confer originating status (i.e. wholly obtained, from yarn etc.). Annex II of the origin protocols introduced by PECS shows that some rules of origin are a mixture of the above three categories of rules.
To measure EU-RoO, we use an index developed byCadot et al.(2006) which is based on Annex II introduced by PECS. This RoO index applies to final goods classified at the HS6 level, and varies from 1 to 7, with 7 representing the most restrictive EU-RoO. Therefore, it is based on change in tariff classification (CTC), and almost always accompanied by other criteria to be met to obtain origin. As a matter of comparison, less than 17% of EU-RoO are uniquely based on CTC, against 89% of NAFTA RoO. Additionally, EU-RoO are largely based on value added requirements (VA). In fact, 26% of tariff lines are based on CTC and VA requirements, 13% of which uniquely rely on VA criteria.8 This seems to suggest that the EU-RoO are more heterogeneous than the NAFTA RoO. To account for the heterogeneous level of complexity embodied in the EU-RoO, the synthetic RoO index constructed byCadot et al. (2006) seems more appropriate than counting the number of final goods products that use a specific intermediate as Conconi et al. (2018) propose to focus on NAFTA .
To translate the EU-RoO for final goods in terms of intermediate goods, we combine the previous two data sets, i.e. US IO1997 table and EU-RoO index by Cadot et al. (2006).9 Using these databases, for each intermediate good we create two measures: a simple and a weighted average of the RoO index faced by the associated final goods. We call these measures the IO-RoO Index and the weighted IO-RoO Index respectively. For each in- termediate, the simple average is constructed by considering all final goods to which this intermediate contributes. This measure, called IO-RoO, is thus computed as follows:
IO-RoOj = P
iRoOij
ni (1)
where j and i refer to the intermediate and final good respectively, and RoOij is the RoO index at the level of the final good, i, associated to each intermediate, j. In equation (1) we sum the RoO index, RoOij, across all final goods into which an intermediate j enters, and we divide it by the number of these IO relations, ni. The drawback associated to the measure in equation (1) is that it gives to all products using a specific intermediate the same importance. To account for the heterogeneity in the use of the intermediate, we also construct a weighted measure where we exploit the direct input-output coefficient requirement which assigns a higher weight to final goods served most prominently by an
8For further details on the product specific rules of origin see Table 2 in Cadot et al.(2006).
9Notice that in combining EU-RoO index to US IO1997 table, we do not find a match for 173 tariff lines.
intermediate good. This weighted measure, called weighted IO-RoO, is then computed as:
weighted IO-RoOj =X
i
RoOij drij P
idrij
(2) wheredrij refers to the direct requirement coefficient.10 This weighted IO-RoO Index is our preferred measure. We believe that it better reflects the production process common across countries and therefore, it provides a more accurate proxy of the production technology.11
Table1presents descriptive statistics for the I-O relations according to the restrictiveness of the RoO index faced by final goods according to the sector of the intermediate goods and the resulting sector-level averages of our IO-RoO Indexes. Columns 1 to 7 present the share of intermediate goods hit by these sourcing restrictions.12 For example, 5.9% of chemical intermediate goods contributes to a final good that faces a RoO Index = 1, with the vast majority of chemicals, 41%, contributing to the production process of a final good facing a RoO Index = 5. Finally, columns 9 and 10 in Table 1 provide the resulting simple and weighted IO-RoO Indexes when looking at the sector of the intermediate good using the 2-digit level of the HS (HS2).
To further stress the importance of considering the synthetic index rather than counting the IO relations, Tables 11and 12 in AppendixA show descriptive statistics at the HS2 on the number of IO relationships and on the average RoO Index faced by intermediates entering into final goods. More precisely, Table 11 calculates the percentage of IO-relations at the HS2 level. For example, about 20% of IO relations are between intermediate goods classified as chemicals entering into the production of final goods classified as chemicals. Conversely, in Table12we compute the simple average RoO at the HS2 level. For example, intermediate goods classified as chemicals face an average RoO index associated to the production of final goods classified as chemicals of 3.3. Comparing Tables 11 and 12 suggests that in the EU context what matters is not only the number of lines into which the intermediate enters;
instead what seems crucial is the restrictiveness of the RoO that the final goods face. For instance, from Table 11we see that 40.7% of intermediate goods classified as textiles enter into the production process of textile goods, and these face an average RoO-Index of 6 (see Table 12). Meanwhile, 17.3% of intermediate goods classified as plastic and rubbers enter into the machinery production process, which faces an average RoO-Index of 5, and fewer of them (13.3% ) enter into the textile production process, which faces a higher RoO-Index.
Tables 11and 12underline that the sourcing restrictions strongly depend on the production process in which the intermediate is used.
10The direct requirement matrix is taken fromConconi et al.(2018).
11A simple average will provide equal weights to final goods served by an intermediate good, even if an intermediate good may, for example, be used only for 0.1% of a final good process.
12The total amount of input-output pairs (amount of intermediates entering in final goods) contained in our data set is 9,499,930.
Table1:DescriptiveStatisticsonEU-RoOindex HS2SectorRoO=1RoO=2RoO=3RoO=4RoO=5RoO=6RoO=7TotalIO-RoOweightedIO-RoO AnimalProducts30.949.716.923.278.31003.53.1 Chemicals5.93.816.422.2413.671004.33.9 Foodstuffs25.73.817.517.515.112.97.51003.63.7 Footwear/Headgear9.338.921.836.47.613.11004.55 Machinery/Electrical5.53.113.918.5493.76.41004.44.7 Metals4.82.914.320.748.93.25.21004.44.5 MineralProducts5.53.819.721.540.93.15.61004.23.8 Miscellaneous6.5314.617.944.84.98.41004.44.5 Plastic/Rubbers73.313.522.341.84.181004.34.3 RawHides,Skins,Leathers9.34.112.418.733.17.4151004.44.1 Stone/Glass3.83.119.418.947.22.45.21004.34.3 Textiles4.61.48.420.637.47.520.11004.95.6 Transportation9.43.412.419.4454.26.21004.24.6 Vegetables19.23.924.816.118.910.36.81003.73.1 WoodProducts6.93.614.220.741.94.48.21004.34.3 Notes:Columns1to8reporttheshareofintermediategoodsthatenterfinalgoodswhichfaceaRoOIndex,from1to7,asashareoftotalnumberofIOrelations.Columns9and10presentmean valuesforsimpleandweightedIO-RoOIndex.
3.2 Tariffs and Imports Data
In the empirical analysis, we consider the effect of PECS on changes in Spokes’ import flows from third countries (RoW) relative to the other Spokes. Then, we further assess the role of diagonal cumulation in relaxing the restrictiveness of RoO using alternative type of comparisons. For this purpose, we use trade data from the World Integrated Trade Solution (WITS) for 1995 and 2002, which correspond to pre- and post-PECS years. Using these years also allows us to exploit the available tariffs information.13
Our group of importing countries is composed of the BAFTA and CEFTA members.
After eliminating those countries with missing tariff observations for the pre-PECS period, i.e. Bulgaria, the Slovak Republic and Slovenia, we are left with 7 Spoke countries: Czech Republic, Estonia, Hungary, Lithuania, Latvia, Poland and Romania. Depending on the specification used, the set of exporting countries is divided into non-participating countries (RoW), EU15 countries (treated as a single group), and Spoke countries (treated also as a group). RoW includes countries with which our Spokes did not have free trade agreements over the period considered. Tables 9 and 10 in Appendix A provide the list of countries included in our study. In the period under study, members of EFTA (Iceland, Norway, and Switzerland) and Turkey (for the subset of industrial goods) also joined the PECS. To assess the impact of diagonal cumulation on a more homogeneous group of countries, our benchmark regression does not include EFTA and Turkey in the set of Spokes exporting countries. Nevertheless, these will be included in robustness checks.14 Table 2 presents de- scriptive statistics of Spoke countries’ imports from RoW and among themselves. Looking at total amounts, Spoke countries’ imports from non-participating countries (RoW) increased by 77%, and by 128% from EU15. The increase in imports among Spoke countries is much larger, at about 161%.15
In the empirical specification, we perform three sets of triple difference estimations. In the main specification, our dependent variable captures product-level changes in each Spoke country’s imports from RoW relative to imports from the Spokes themselves, ∆impj,srow−
∆impj,ss. In the second specification, our dependent variable captures product-level changes in each Spoke country’s imports from RoW relative to imports from the EU15, ∆impj,srow−
∆impj,sEU15. Finally, in the last specification the dependent variable captures product-level changes in each Spoke country’s imports from the other Spokes countries relative to imports from the EU15, ∆impj,ss−∆impj,sEU15.
In our data set, imports from RoW retain the by-product-exporter-importer dimension, i.e. imports by Estonia at the HS6 level from each third country. Meanwhile, imports of each Spoke from the other Spokes lose the exporter-importer dimension and are aggregated at the HS6 level, i.e. total imports into Estonia of a particular HS6 line from all the other 6 Spokes (the same is true for imports from EU15). The rationale behind this aggregation choice is to capture the third country effects compared to the cumulation zone effect. Furthermore,
13For data availability reasons we cannot take a symmetric 5-years window around 1997. More details on tariffs data are provided below.
14Notice that we never include EFTA and Turkey in our group of importing countries.
15Table16in AppendixApresents descriptive statistics about Spokes’ imports from EU15.
we only keep imports from non-participating countries or from Spoke countries that have at least a non-zero entry, i.e. we eliminate imports from the RoW or from other Spoke countries that take the value zero in both years. This implies that, first the dependent variable does not take the value zero, unless changes in imports flows from the Spoke countries equals change in imports from a third country for a specific product defined at the HS6. Second, this approach implies that we are comparing changes in trade flows among HS6 products for which Spoke countries compete with the rest of the world.
We combine trade data with tariff-level data from UNCTAD Trade Analysis Information System (TRAINS), which provides HS-based tariff line level (HS 6-digit). When tariff data are missing, we use the nearest data point available favoring earlier years when possible (both for pre- and post-PECS). Our tariff measure captures by how much Spoke countries lowered tariffs on imports among themselves compared to tariffs on products from non-participating countries or from EU15. We compute the tariff change using both preferential tariffs and most favored nation (MFN) tariffs. Specifically, we use MFN tariff every time that we have a missing preferential tariff. Table 2presents descriptive statistics on tariffs applied by Spoke countries with respect to other Spokes and RoW countries.16 Notice that the change in import tariff from RoW, ∆τj,srow, keeps a bilateral dimension (at the six-digit level of the HS), while the change in tariff from the other Spokes, ∆τj,ss, and EU15 countries, ∆τj,sEU15, are constructed as a simple average of the HS6 tariff applied by each Spoke country to the other Spokes, and EU15 countries.
16Table16in AppendixApresents descriptive statistics about Spokes’ import tariff applied to EU15.
Table2:DescriptiveStatisticsonImportsandTariffs Insampleaverages:AvgSpokesimportsfromRoWAvgSpokesimportsfromSpokesAvgTariffsappliedtoRoWAvgTariffsappliedtoSpokes PrePECSPostPECSPrePECSPostPECSPrePECSPostPECSPrePECSPostPECS AnimalProducts152,79224,44180,00314,888,6010,498,7511,80 Chemicals114,13183,74698,031201,985,604,734,574,11 Foodstuffs200,35255,67683,801739,3814,8919,4914,7817,42 Footwear/Headgear59,27129,61339,72791,9110,139,629,466,02 Machinery/Electrical90,37342,53394,251077,376,384,555,543,76 Metals69,54141,12529,70958,775,985,625,544,73 MineralProducts3511,166073,353674,9312659,992,231,911,751,91 Miscellaneous46,5094,55207,30569,637,155,385,953,94 Plastic/Rubbers73,46129,95858,211775,217,966,456,865,41 RawHides,Skins,Leathers63,48103,78201,82518,697,777,217,214,66 Stone/Glass30,6562,25337,48668,087,586,656,885,53 Textiles38,1872,11164,20395,9010,1310,249,657,21 Transportation185,43532,021234,723244,347,266,546,285,46 Vegetables137,82129,08345,75353,716,508,156,037,62 WoodProducts66,75109,40846,921728,526,645,395,774,33 Notes:ValuesareinthousandsofUS$.Alltariffsareexpressedinpercentageterms.Thefirsttwocolumnsrefertoaverageimportsfromnonpartnercountries,withwhichourSpokesdidnothavefree tradeagreement(FTA)duringoursampleperiod.
4 Empirical Strategy
4.1 Identification Strategy
We provide an empirical assessment of the impact of relaxing RoO on sourcing decisions of intermediate goods, and therefore its role in shaping international supply chains. We employ the introduction of the PECS, which introduced diagonal cumulation, and focus on peripheral countries like BAFTA and CEFTA countries (the Spokes) to investigate their changes in imports of intermediate goods between 1995 and 2002.
The empirical analysis is based on a triple difference estimation in the spirit ofConconi et al.(2018). In our benchmark specification we compare changes in imports of intermediate goods by each Spoke country from countries outside the PECS cumulation zone (RoW countries) and from the Spokes themselves. The change is captured by considering pre- and post-PECS periods. Since PECS was introduced in 1997 and all Spoke countries considered in our analysis joined the PECS system by 1999, we look at changes in their imports between 1995 and 2002. Specifically, let us define:
∆impj,srow =α0+α1IO-RoOj +α2∆τj,srow+Xj +γsrow +j,srow (3) where ∆impj,srowis the change in log imports of intermediate goodj, defined at the six-digit level of the HS, by each Spoke country from the RoW between 1995 and 2002. ∆τj,srow is the change in log tariffs with respect to the RoW, andXj are product level trends.17 Finally γsrow are bilateral time-variant fixed effects.
To capture the change in imports of intermediate goods among Spoke countries, we define:
∆impj,ss =β0 +β1IO-RoOj +β2∆τj,ss+Xj +γs+j,ss (4) where ∆impj,ss is the change of log imports in intermediate good j, defined at the six-digit level of HS, by each Spoke country from all other Spoke countries between 1995 and 2002.
IO-RoOj is our our proxy for the effect of PECS (related to the restrictiveness of the RoO facing each intermediate good). ∆τj,ss is the change in the HS6 average tariffs applied by each Spoke country to the other Spokes. Finally, Xj are product-level trends, andγs is the time-variant importer fixed effect.
Subtracting equation (4) from (3) yields our benchmark specification:
∆impj,srow−∆impj,ss =γ0+γ1IO-RoOj +γ2∆τj+γsrow+γs+j,srows (5) where ∆τj = ∆τj,srow − ∆τj,ss captures by how much Spoke countries lowered tariffs on imports among themselves compared to tariffs on products from the RoW.18Since we assume that product-level trends are the same for imports from RoW and from Spoke countries,
17As explained in section3.2, our tariff variable is computed using both preferential and MFN tariffs. In particular, we use the MFN tariff every time we have a missing preferential tariff.
18The change in tariff is defined as ∆τj,srow−∆τj,ss, where ∆τj,srow = log(1 +τj,srow2002)−log(1 + τj,srow1995), and ∆τj,ss=log(1 +τj,ss2002)−log(1 +τj,ss1995).
Xj cancels out in equation (5).19 IO-RoOj is our proxy for the effect of PECS. Since the impact of diagonal cumulation is linked to the restrictiveness of the RoOs, PECS is captured by the restrictiveness of the IO-RoO facing each intermediate good (simple and weighted average). The IO-RoO index captures the effect of diagonal cumulation rather than the effect of RoOs.20
Our identification strategy relies on the triple difference specification where we compare changes in imports, before and after PECS, from third countries and from the other Spoke countries. Specifically, our treatment, PECS, is captured by the second part of the triple difference, which considers imports from those peripheral countries signing PECS (which could diagonally cumulate). Then, using this econometric specification we study how PECS (captured by IO-RoOj) differently affect imports from RoW and Spokes. The use of a triple difference, as in equation (5), allows us to reduce omitted variable bias compared to the standard estimates used in the literature. Similar to the standard approach generally used to study the effect of free trade agreements on trade flows, our approach reduces potential bias deriving from unobservable time-invariant product characteristics. Additionally, by taking a triple difference, we are able to control for unobservable time-variant product characteristics (product-level trends).21 In every specification, we correct for heteroskedastic standard errors using the Eicker-Huber-White correction.22
The variable IO-RoOj captures the effect of diagonal cumulation. This variable should affect trade flows of intermediate good j depending on which final good the intermediate goodj contributes to, and on the level of RoO restrictiveness this final good faces. We expect the sign of IO-RoOjto be negative if diagonal cumulation led Spoke countries to import more intermediate goods from other Spoke countries relative to the non-participating countries (RoW). Additionally, we expect ∆τj to be negative since it captures by how much Spoke countries lowered tariffs on imports among themselves compared to tariffs on products from the RoW.23
Owing to the nature of the experiment, we further assess the role of diagonal cumulation in relaxing the restrictiveness of RoO using alternative control groups over the same period,
19More specifically, our dependent variable is ∆impj,srow −∆impj,ss, where ∆impj,srow = log(1 + Impj,srow2002)−log(1 +Impj,srow1995), and ∆impj,ss=log(1 +Impj,ss2002)−log(1 +Impj,ss1995).
20Notice that our IO-RoO index is constructed using the PECS rules of origin protocols (seeCadot et al., 2006). The set of RoO included in the PECS protocols do not represent a fundamental change with respect to those used in bilateral agreement signed by the EU with each CEECs countries in pre-PECS period (see Estevadeordal and Suominem,2006andBart and Graafsma,1999).
21Exploiting the panel structure would have required HS6 fixed effects. But this implies dropping our measure IO-RoOj, since it only varies at HS6, and not over time.
22FollowingAbadie et al.(2017), we think our design problem does not face aggregation issues that would require clusters. First, our sampling does not follow a two-stage process, and second PECS is at the country level and affect the universe of all the importing countries under analysis. Additionally, it is important to notice that, despite RoO are defined mostly at the 4 digit levels and apply to final goods, our estimation strategy focuses on intermediate goods. In fact, our variable of interest, IO-RoO, is constructed by: i) matching the input-output table with the RoO variables applied to each final good, ii) and taking averages by intermediate goods. This implies that our IO-RoO measure has variation within HS4.
23We also run additional regressions where we change the group of Spokes’ exporting countries used to create ∆impj,ss. We first consider changes in imports coming only from Spoke countries belonging to the same FTAs (either BAFTA or CEFTA). Then, we consider changes in imports coming from Spoke countries outside the FTA (i.e. imports of each BAFTA country from CEFTA countries, and the reverse). Results are all consistent with our benchmark findings.
i.e. 1995-2002. This will allow us to better disentangle trade creation from trade diversion effects in intermediate goods. Firstly, we estimate the triple difference comparing changes in imports by each Spoke country from the RoW to those from the EU15, ∆impj,srow −
∆impj,sEU15. This yields:
∆impj,srow−∆impj,sEU15=γ0+γ1IO-RoOj +γ2∆τj+γsrow+γs+j,srowEU15 (6) Then, we compare changes in imports by each Spoke country from the rest of the Spoke countries to changes in imports from the EU15, ∆impj,ss−∆impj,sEU15. This gives:
∆impj,ss−∆impj,sEU15 =γ0+γ1IO-RoOj+γ2∆τj +γsEU15+γs+j,ssEU15 (7) A positive coefficient of IO-RoOj in both control groups would indicate that producers in Spoke countries are importing less from the EU15 compared to RoW and the other Spoke countries. This could happen if, following diagonal cumulation, producers in the PECS cumulation zone were able to reassess sourcing choices more in line with their cost minimization objectives, and still obtain preferential access for exports of their final good.
4.2 Threats to Identification
A key econometric aspect when analyzing the effect of trade policy is that trade policy can result from economic conditions. For example, trade policy can ratify changes in trade that were already happening for other reasons (Goldberg and Pavcnik, 2016). The endogeneity of trade policy generates bias estimates. And according to the specific case considered, this would overstate or understate the effect of trade policy on trade flows. It may understate if the policy change simply follow trade changes that have already occured. Alternatively, the effect of trade policy on trade flows may be overstated if countries expect increases in trade and as a result form a trade agreement (Goldberg and Pavcnik, 2016). In this latter case, the normative implications of the findings may not be generalized.
In relation to our paper, recent studies have highlighted that the fragmentation of pro- duction process generates incentive for firms to influence trade policy (Blanchard, 2007, Blanchard et al.,2016, andBlanchard and Matschke,2015). In Europe, after the fall of the iron curtain, a process of unbundling of the production process took place: parts and com- ponents crossing borders many times before being sold to consumers (see Baldwin, 2013).
However, the economic integration of Europe’s markets were opposed by the complex net- work of about 60 bilateral and plurilateral free trade agreement regulating European trade.
This plethora of FTAs was characterized by RoO with bilateral cumulation, which led to the spaghetti bowl pattern. EU producers, in particular when delocated in other Spoke countries, were hurted in two ways. First, bilateral cumulation hindered the EU firms located in the Spoke economies from sourcing their inputs most efficiently. Second, the different set of RoO associated to different FTAs imposed additional administrative costs.
To solve the drawbacks associated with overlapping of FTAs, policymakers implemented the Pan-European Cumulation System which instituted common rules of origin, and diagonal
cumulation of origin rules. PECS allowed EU affiliates to enjoy more efficient combination of intermediate inputs, while still access the EU market under preferential tariff.
In this paper, we take the perspective of several peripheral European countries (Spokes), where producers played a less active role towards PECS. Additionally, the PECS deal was part of broad process of European integration, motivated mainly by political factors. Thus, the issue related to endogeneity of trade policy should be less severe. Yet econometric endogeneity may still be present due to pre-existing trends. Pre-existing trends might be related to the increasing economic integration, and can be spuriously correlated with the trade policy changed considered. To rule out this possibility, in Section 5.2 we implement the triple difference methodology using different pre-PECS periods, i.e. 1992-1993, and 1992-1996. Due data availability problems, related to import flows and tariffs, we focus uniquely on Hungary.
5 Empirical Results
The following sections present the results and some robustness and sensitivity analysis.
5.1 Benchmark Specification
Table 3 reports OLS estimates of equation (5), which compares changes in imports by each Spoke country from countries outside the PECS cumulation zone (RoW) to those from the rest of the Spoke countries. Columns (1) and (3) include only the IO-RoO Indexes (weighted and simple) controlling for importer and for exporter time-variant characteristics. Columns (2) and (4) include the differences in the change of tariffs, as described in Section 3.2.
The coefficients of both the simple and weighted average IO-RoO indexes are negative and significant. This suggests that diagonal cumulation reduced imports of intermediate goods of Spoke countries from non-participating countries relative to the other Spoke countries that joined the cumulation zone in the period under analysis. Our preferred measure, weighted IO-RoO, has the largest coefficient. The coefficient of the change in tariff, ∆τj, is always negative and significant. This confirms that the change in preferential tariff reduced Spokes imports from non-participating countries compared to imports from the other Spokes.
Table 3: PECS and change in imports from RoW and Spokes
Dep Var: Change in log Imports: RoW and other Spokes
(1) (2) (3) (4)
weighted IO-RoOj -0.116*** -0.138***
(0.008) (0.016)
IO-RoOj -0.071*** -0.087***
(0.012) (0.021)
∆τj -3.776*** -3.889***
(0.652) (0.658)
Observations 117,877 25,890 117,877 25,890
R-squared 0.077 0.115 0.075 0.113
Importer FE Yes Yes Yes Yes
Exporter FE Yes Yes Yes Yes
Notes: OLS estimation. The dependent variable is the difference between changes in log imports of intermediate j from non- participating countries between 1995 and 2002, and the corresponding change of imports from the rest of Spoke countries,
∆impj,srow−∆impj,ss. ∆τjis the change in preferential tariff (where we use the applied MFN in case of missing preferential tariff information). IO-RoOjand weighted IO-RoOjrepresent our simple and weighted average measures of the restrictiveness in RoO respectively. Importing countries include: Czech Republic, Estonia, Hungary, Lithuania, Latvia, Poland, and Romania.
Tables4and 5propose OLS estimates using the alternative groups of countries. Table4 shows changes in imports of intermediate j of each Spoke country from the RoW compared to changes in imports from the EU15, as in equation (6). Similarly to Table 3, columns (1) and (3) include only our IO-RoO Index (weighted and simple), controlling for importer and exporter time-variant fixed effects. Columns (2) and (4) include the differences in the change of tariffs. The coefficients of both the simple and weighted average IO-RoO indexes are positive and significant, with the exception of the simple average IO-RoO index which loses significance when tariffs are not controlled for. These results suggest that diagonal cu- mulation increased imports of intermediate goods of Spoke countries from non-participating countries relative to the EU15. The mechanism behind this result relates to the fact that RoO in general do not require that products must be fully obtained within the cumulation zone to qualify for preferential access. Therefore, diagonal cumulation could have enlarged the set of countries from which it was possible to cumulate inputs for obtaining originating status. Specifically, it may have led exporting firms in Spoke countries to reassess sourcing decisions in favor of the RoW, rather than the EU15. Concerning the coefficient of the change in tariff, ∆τj, this is always negative and significant. This suggests that the change in preferential tariff reduced Spoke’s imports from the rest of the world compared to the increase in preferential access provided by the EU15.
Table 4: PECS and imports from RoW and EU15
Dep Var: Change in log Imports: RoW and EU15
(1) (2) (3) (4)
weighted IO-RoOj 0.044*** 0.031**
(0.007) (0.013)
IO-RoOj 0.009 0.035*
(0.010) (0.018)
∆τj -1.747*** -1.725***
(0.317) (0.317)
Observations 127,960 31,720 127,960 31,720
R-squared 0.104 0.163 0.103 0.163
Importer FE Yes Yes Yes Yes
Exporter FE Yes Yes Yes Yes
Notes: OLS estimation. The dependent variable is the difference between changes in log imports of intermediatejfrom non- participating countries between 1995 and 2002, and the corresponding change of log imports from the EU15, ∆impj,srow−
∆impj,sEU15. ∆τj is the change in preferential tariff (where we use the applied MFN in case of missing preferential tariff information). IO-RoOj and weighted IO-RoOjrepresent our simple and weighted average measures of the restrictiveness in RoO respectively. Importing countries include: Czech Republic, Estonia, Hungary, Lithuania, Latvia, Poland, and Romania.
In sum, diagonal cumulation appears to have also allowed producers to rethink sourcing decisions more in line with costs minimization factors. Before the introduction of PECS, to obtain preferential access for their final good in the EU15 market, producers in peripheral countries could cumulate inputs only accordingly to the specific FTA signed. Following PECS, producers could cumulate inputs from a larger set of countries to fulfill RoO, and still enjoy preferential access to the EU15. This may have led producers to change the composition in terms of origin of their input basket and still obtain preferential tariffs.
This mechanism seems also confirmed in Table5, where we consider the impact of diago- nal cumulation on changes in imports of each Spoke country from the other Spoke countries compared to changes in imports from the EU15 (equation (7)). In columns (1) and (2) we use the weighted average IO-RoO measure, which assigns a higher weight to final goods most prominently served by intermediates. The positive and significant coefficient suggests that PECS have led exporting firms in Spoke countries to reassess sourcing decisions in favor of the other Spoke countries in the PECS network, rather than the EU15. Differently, in columns (3) and (4) we use the simple average IO-RoO Index, which assumes that an intermediate good is used in similar proportions across every final product to which the intermediate good contributes to. The IO-RoO Index coefficient is negative, but only when we do not control for the effect of tariff changes. The coefficient associated to the change in tariff, ∆τj, captures the effect of the process of economic integration in the European area (by comparing changes in tariffs among spokes to changes in tariffs to the EU). The sign is always positive but not significant. This suggests that the process of tariffs reduction had homogenous effects in the whole European area.
